Limeglass Fusion Search for AI – Enhances Existing Search Engines
Improve the Quality of Search & AI Retrieval Results
Current challenges when searching Financial Content
Tuning search engines to work effectively with investment research or other financial content is a significant and often underestimated challenge.
Despite using a top-quality search engine, adhering to best practices, and embracing advanced methods like semantic search, chunking, and vector embeddings, search engine results often still fall short.
At Limeglass, we are the financial content discovery technology experts, and understand that an effective solution only comes from overcoming the core challenges:
- Time Recency Trade-off: Search engines are not tuned to respect time recency, often returning ‘relevant’ content that is out of date.
- Missing Context: Single documents cover a wide variety of themes, naïve chunking strategies miss crucially important context.
- Understanding Financial Language: Without a well-defined and structured understanding of financial domain knowledge, important results are missed.
- Awareness of Subject Matter Experts: Without domain knowledge, search engines are unable to adequately promote content produced by the subject matter experts.
- False Positives & False Negatives: Due to noise and information loss, search results contain non-relevant results and miss relevant content.
With the advent of GenAI, LLMs, and in particular Retrieval Augmented Generation (RAG) – the search retrieval quality has never been more important, it is the biggest factor impacting the quality of an LLMs output.
Limeglass Fusion Search for AI optimises your existing search engine for Financial Content
Limeglass Fusion Search for AI is designed to tackle the challenges of searching investment research and more.
By integrating the power of Graph and Hybrid Search into existing search engines and leveraging the Limeglass Knowledge Graph of over 200,000 topics, it achieves substantially better results in clients’ tests and benchmarks.
Fusion Search for AI provides two primary API components that enhance the existing search stack:
Fusion Search for AI – Index Pre-Processor
Pre-processes financial research content before index ingestion.
Fusion Search for AI Index Pre-Processor API ensures deep granular labeling, advanced contextual chunking and metrics at both document and chunk levels, resulting in standardized metadata across all content ingested.
Providing a detailed and consistent indexing of content, while supplementing the publisher’s own metadata extensions and RIXML files (which can have variable quality).
Fusion Search for AI – Query Parser
Interprets the user’s search / prompt to enhance the search engine query.
The Fusion Search for AI Query Parser API interprets the context of the question using the Limeglass Knowledge Graph. Generating Graph / Hybrid powered search queries, including rich metadata to ensure more accurate retrieval from the index.
How does Fusion Search for AI Index Pre-Processor improve ingestion & chunking?
Providing more accurate search results
Issues with current methods of ingesting Financial Market Research content into a Search Engine Index
If the content in your search engine index is suboptimal, any results will be disappointing
Fusion Search for AI ensures you do not lose important contextual information
Contextual Search using Content Atomisation™, Document & Knowledge Graphs
Fusion Search for AI:
- retains Context in a Document Graph
- removes noisy sections
- adds Domain Knowledge via the Limeglass Knowledge Graph
- better respects Time Recency
- provides Contextual Chunking which:
- improves Full-Text & Vector Embeddings
- preserves chunk relationships
- adds Graph Tags at document, contextual & chunk-level
How does Fusion Search for AI Query Parser improve search query parsing?
Ensuring more accurate retrieval
Fusion Query Parser enhances search queries with metadata
Fusion Query Parser parses the original query adding metadata to ensure more accurate retrieval
- Decomposes Search Query using Limeglass Knowledge Graph
- Identifies key Semantic Graph Tag Concepts
- Suggests additional Contextual Filters – Promotes authoritative content
- Re-writes question into Search Text using LLM (optional)
- Disambiguates Ambiguous Terms – Identifies alternative interpretations
- Additional Semantic Disambiguation using LLM (optional)
- Uses Tags as Filters & Sparse-Vectors within search to better respect Time Recency
- Generates Search Engine query template
Get in touch to book a demo or find out more
If you would like to speak with the team or book a demo to see how our technology can empower your team, please get in touch using the form below: